Download Asymptotics of repeated interaction quantum systems Laurent Bruneau , Alain Joye

Asymptotics of repeated interaction
quantum systems
Laurent Bruneau∗, Alain Joye†‡, Marco Merkli§
November 4, 2005
Abstract
A quantum system S interacts in a successive way with elements E of a chain of
identical independent quantum subsystems. Each interaction lasts for a duration
τ and is governed by a fixed coupling between S and E. We show that the system,
initially in any state close to a reference state, approaches a repeated interaction
asymptotic state in the limit of large times. This state is τ –periodic in time and
does not depend on the initial state. If the reference state is chosen so that S
and E are individually in equilibrium at positive temperatures, then the repeated
interaction asymptotic state satisfies an average second law of thermodynamics.
1
Introduction
In this introduction we outline our main results and the relevant ideas of their proofs.
Suppose a quantum system S interacts with another one, E, during a time interval
[0, τ ), where τ > 0 is fixed. Then, for times [τ, 2τ ), S interacts in the same fashion with
another copy of E, and so on. The assembly of the systems E (which are not directly
coupled among each other) is called a chain, C = E +E +· · · . The system S +C, with an
interaction as described above, is called a repeated interaction quantum system. One
may think of S as being the system of interest, like a particle enclosed in a container,
and of C as a chain of measuring apparatuses E that are brought into contact with the
particle in a sequential manner.
Our goal is to study the large time behaviour of repeated interaction quantum
systems, and in particular, to describe the effect of the repeated interaction on the
system S. One of our main results is the construction of the time–asymptotic state,
which we call a repeated interaction asymptotic state (RIAS).
States of S and E are represented by vectors (or density matrices) in the Hilbert
spaces HS and HE , respectively. We assume that dim HS < ∞, while dim HE ≤ ∞. The
∗
CPT-CNRS, UMR 6207 Université du Sud, Toulon-Var, BP 20132, F-83957 La Garde Cedex,
France. Email: [email protected]
†
Institut Fourier, UMR 5582, CNRS-Université de Grenoble I BP 74, 38402 Saint-Martin d’Hères,
France. Email: [email protected]
‡
Laboratoire de Physique et Modélisation des Milieux Condensés, UMR 5493, CNRS-Université de
Grenoble I, BP 166, 38042 Grenoble, France
§
Department of Mathematics and Statistics, McGill University, 805 Sherbrooke Street West, Montreal, QC, H3A 2K6 Canada. Email: [email protected], http://www.math.mcgill.ca/emerkli/
BJM November 4, 2005
2
observables of S and E are bounded operators, they form the (von Neumann) algebras
MS ⊂ B(HS ) and ME ⊂ B(HE ). Observables evolve according to the Heisenberg
dynamics t 7→ τSt (AS ) and t 7→ τEt (AE ), where τSt and τEt are groups of ∗automorphisms
of MS and ME , respectively.
We assume that there are distinguished “reference” states, represented by the vectors ΩS ∈ HS and ΩE ∈ HE , and for the purposes of the introduction, we shall take
ΩS , ΩE to be equilibrium states with respect to τSt , τEt , for inverse temperatures βS , βE ,
respectively. It is useful to pass to a description of the dynamics of vectors in HS , HE
(Schrödinger dynamics). There are selfadjoint operators LS , LE , called the standard
Liouville operators, uniquely specified by
t
τ#
(A) = eitL# Ae−itL# , and L# Ω# = 0,
(1.1)
where # stands here for either S or E.
The Hilbert space of the entire system is given by H = HS ⊗ HC , where HC , the
Hilbert space of the chain, is the infinite tensor product ⊗m≥1 HE . The non-interacting
dynamics is defined on the algebra MS ⊗m≥1 ME by τSt ⊗m≥1 τEt .
We consider interactions of the following kind. Fix an interaction time τ > 0.
During the interval [0, τ ), S interacts with the first element E in the chain C, while
all other E’s evolve freely. The interaction is specified by an operator V ∈ MS ⊗ ME .
In the next time interval, [τ, 2τ ), S interacts with the second element in the chain,
through the same interaction operator V , and all other elements evolve freely, and so
on. For t ≥ 0 we set
t = m(t)τ + s(t),
(1.2)
where m(t) is the integer measuring how many complete interactions of duration τ have
taken place at the instant t, and where 0 ≤ s(t) < τ . We define the repeated interaction
(Schrödinger) dynamics, for t ≥ 0, ψ ∈ H, by
URI (t)ψ = e−is(t)Lm(t)+1 e−iτ Lm(t) · · · e−iτ L1 ψ,
e
e
e
(1.3)
where
e m = Lm +
L
X
LE,k
(1.4)
k6=m
is the generator of the total dynamics during the time interval [(m − 1)τ, mτ ). We have
introduced Lm , the operator on H that acts trivially on all elements of the chain except
for the m–th one, and which, on the remaining part of H (which is just HS ⊗ HE ), acts
as
L = LS + LE + V.
(1.5)
In (1.4), LE,k denotes the operator on H that acts nontrivially only on the k–th element
of the chain, on which it equals LE .
A state ω given by a density matrix on H is said to be normal. Our goal is to
understand the time-asymptotics (t → ∞) of expectations
t
ω URI (t)∗ OURI (t) ≡ ω αRI
(O) ,
(1.6)
BJM November 4, 2005
3
for normal states ω and for certain classes of “observables” O. As mentioned above, we
may regard S as the system of interest, so we certainly want to treat the case O ∈ MS .
Another type of physical observable is of interest as well. Imagine we want to measure
the variation, say, of the energy of S at a certain moment in time. This measuring
process involves the system S, but also the element of the chain that is in contact
with S at the given moment. We call such an observable an instantaneous observable.
Various generalizations can be considered, see Section 2, but we limit our discussion in
this introduction to the two kinds of observables just described.
Asymptotic state. Let O be an instantaneous observable, determined by AS
and AE . This means that at time t = m(t)τ + s(t), (1.2), O measures AS ⊗ AE on
the system S + E, where E is the (m(t) + 1)–th element in the chain C. We show in
Theorem 2.3 that, under a natural assumption on the interaction, we have
s(t)
t
(O) − ω+ P αRI (AS ⊗ AE )P −→ 0,
(1.7)
ω αRI
as t → ∞, where ω+ is a state on MS which does not depend on ω (c.f. (1.13)), and
where
P = 1lHS ⊗m≥1 PΩE ,
(1.8)
with PΩE denoting the orthogonal projection onto CΩE . We identify the range of P
with HS . Relation (1.7) shows that the expectation of an instantaneous observable in
any normal initial state approaches a τ –periodic limit function (t 7→ s(t) is τ –periodic).
The speed of convergence in (1.7) is exponential, ∼ e−tγ/τ , where γ > 0 is a constant
depending on the interaction.
The restriction of the RIAS to the algebra of instantaneous observables characterized by AS ∈ MS and AE ∈ ME is the τ –periodic state
s(t)
AS ⊗ AE 7→ ω+ P αRI (AS ⊗ AE )P
(1.9)
on MS ⊗ME , see (1.7) (and (2.26) for the definition of the RIAS acting on more general
observables).
The above–mentioned assumption on the interaction is an ergodicity assumption on
the dynamics reduced to the system S. More precisely, we construct a (non–symmetric)
operator K on H s.t.
t
P αRI
(AS )P ΩS = (P eiτ K P )m(t) P eis(t)K AS P ΩS ,
(1.10)
P eiτ K P ΩS = ΩS .
(1.11)
with the property
We assume that 1 is a simple eigenvalue of P eiτ K P , and that all the other eigenvalues
lie strictly inside the complex unit disk. (We prove in Section 3 that this holds for
concrete models). As a consequence of this assumption, we have
(P eiτ K P )m(t) −→ |ΩS ihΩ∗S |,
(1.12)
as t → ∞, where Ω∗S is the unique vector in HS satisfying (P eiτ K P )∗ Ω∗S = Ω∗S and
hΩ∗S , ΩS i = 1. To arrive at result (1.7), where
ω+ (·) = hΩ∗S , · ΩS i ,
(1.13)
BJM November 4, 2005
4
we use (1.12) together with an argument involving a cyclicity property of ΩS .
Our approach is constructive in the sense that the asymptotic characteristics of the
system, such as the state ω+ , the speed of convergence γ, and the asymptotic dynamics
in (1.7) can be calculated by rigorous perturbation theory (in V ).
Correlations & reconstruction of initial state. The (time dependent) asymptotic expectations of observables do not depend on the initial state of the system, c.f.
(1.7). However, asymptotic correlations do, and together with the asymptotic expectations they permit to reconstruct the intial state in the following way.
Take a normal state ω, an instantaneous observable O determined by AS ∈ MS ,
AE ∈ ME , and an observable A ∈ M. We show in Theorem 2.5 that
s(t)
t
ω AαRI
(1.14)
(O) − ω(A) ω+ P αRI (AS ⊗ AE )P −→ 0,
as t → ∞ (exponentially fast). According to (1.7) and (1.14), knowledge of the asymptotic correlation function C+ (t; A, O), and of the asymptotic expectation E+ (t; O), determined respectively by
t
lim |ω(AαRI
(O)) − C+ (t; A, O)| = 0 and
t→∞
t
lim |ω(αRI
(O)) − E+ (t; O)| = 0,
t→∞
allows for a reconstruction of the initial state ω according to
ω(A) =
C+ (t; A, O)
.
E+ (t; O)
(1.15)
Energy, entropy, average 2nd law of thermodynamics for RIAS. The
t (L
e m(t)+1 ), where L
e m is given by (1.4), has a well-defined variation
formal quantity αRI
in t. It is not hard to see by explicit calculation that this variation is zero in all time
intervals [(m − 1)τ, mτ ), and that it undergoes a jump
mτ
j(m) = αRI
(Vm+1 − Vm )
as time passes the moment mτ . Here, Vk denotes the operator V acting nontrivially
only on HS and the k–th element HE of the chain Hilbert space HC . We interpret the
variation of the above formal quantity as the (time dependent) observable of variation
in total energy of the system.
We show in Section 2.4.1 that for any normal initial state ω, the variation in energy
during any time interval of length τ takes the asymptotic expectation value ω+ (j+ ),
where
Z τ
τ
s
j+ = P V P − P αRI (V )P = −i
P αRI
([LS + LE , V ])P ds.
(1.16)
0
(Here and in the rest of the paper we understand commutators to be defined in the
form sense, but none of our arguments involve delicate domain questions with regards
to commutators.) We define the (average) asymptotic energy production dE+ to be the
change in energy during any interval of duration τ , divided by τ , in the limit of large
times. This quantity is given by
dE+ =
1
ω+ (j+ )
τ
(1.17)
BJM November 4, 2005
5
and is independent of the initial state ω. We show in Section 2.4.2 that ω+ (j+ ) ≥ 0.
Denote by ω0 the state on M determined by the vector ΩS ⊗m≥1 ΩE ∈ H, and let
Ent(ω|ω0 ) denote the relative entropy of the normal state ω w.r.t. ω0 . We think it
is natural to define the entropy as a non–negative quantity, and our definition of it
differs by a sign from the one given in [5]1 . We define the (average) asymptotic entropy
production dS+ to be the change of (relative) entropy in any interval of duration τ ,
divided by τ , in the limit of large times. We prove in Section 2.4.2 that
dS+ =
βE
ω+ (j+ ),
τ
(1.18)
where βE is the inverse temperature of the elements in the chain. The asymptotic
entropy production does not depend on the initial state ω. We may combine (1.17)
and (1.18) to arrive at an average 2nd law of thermodynamics for repeated interaction
quantum systems,
dE+ = TE dS+ ,
(1.19)
where TE = 1/βE is the temperature of the chain. Relation (1.19) is independent of the
initial state of the system, and it holds for any repeated interaction system (V and τ ).
Repeated interaction quantum systems emerge in quantum optics, see e.g. [11, 4]
and references therein. The effective evolution of the small system in certain specific
regimes of parameters related to the Van Hove limit has been studied in [2]. In some
parameter regimes repeated interaction models can be considered as coarse grained
versions of a system S in contact with quantum noises [3].
2
2.1
Model & Results
Repeated interaction models
The models we consider consist of a system S which is coupled to a chain C = E +E +· · ·
of identical elements E. We describe S and E as W ∗ –dynamical systems (MS , τSt ) and
(ME , τEt ), where MS , ME are von Neumann algebras “of observables” acting on the
Hilbert spaces HS , HE , respectively, and where τSt and τEt are (σ–weakly continuous)
groups of ∗automorphisms describing the Heisenberg dynamics. In this paper, we
consider the situation dim HS < ∞ and dim HE ≤ ∞.
We assume that there are distinguished vectors ΩS ∈ HS and ΩE ∈ HE , determining
states on MS and ME which are invariant w.r.t. τSt and τEt , respectively, and we assume
that ΩS and ΩE are cyclic and separating for MS and ME , respectively. One may
typically think of these distinguished vectors as being KMS vectors.
The Hilbert space of the chain C is defined to be the infinite tensor product
HC = ⊗m≥1 HE
(2.1)
ΩC = ΩE ⊗ ΩE · · · .
(2.2)
w.r.t. the reference vector
1
For a finite system we have Ent(ω|ω0 ) = Tr(ρ(log ρ − log ρ0 )), where ρ and ρ0 are density matrices
determining the states ω and ω0 , respectively, and where ρ0 > 0.
BJM November 4, 2005
6
In other words, HC is obtained by taking the completion of the vector space of finite
linear combinations of the form ⊗m≥1 ψm , where ψm ∈ HE , ψm = ΩE except for finitely
many indices, in the norm induced by the inner product
Q
h⊗m ψm , ⊗m χm i = m hψm , χm iHE .
(2.3)
We introduce the von Neumann algebra
MC = ⊗m≥1 ME
(2.4)
acting on ⊗m≥1 HE , which is obtained by taking the weak closure of finite linear combinations of operators ⊗m≥1 Am , where Am ∈ ME and Am = 1lHE except for finitely
many indices.
The operator algebra containing the observables of the total system is the von
Neumann algebra
M = MS ⊗ MC
(2.5)
which acts on the Hilbert space
H = HS ⊗ HC .
(2.6)
The repeated interaction dynamics of observables in M is characterized by an interaction time 0 < τ < ∞ and a selfadjoint interaction operator
V ∈ MS ⊗ ME .
(2.7)
For times t ∈ [τ (m − 1), τ m), where m ≥ 1, S interacts with the m–th element of the
chain, while all other elements of the chain evolve freely (each one according to the
dynamics τE ). The interaction of S with every element in the chain is the same (given
by V ).
Let LS and LE be the standard Liouville operators (“positive temperature Hamiltonians”, c.f. references of [6, 9]), uniquely characterized by the following properties:
L# (where # = S, E) are selfadjoint operators on H# which implement the dynamics
t ,
τ#
t
τ#
(A) = eitL# Ae−itL# , ∀A ∈ M# ,
(2.8)
and
L# Ω# = 0.
(2.9)
L = LS + LE + V,
(2.10)
We define the selfadjoint operator
omitting trivial factors 1lS or 1lE (by LS in (2.10) we really mean LS ⊗ 1lE , etc). L
generates the automorphism group eitL · e−itL of MS ⊗ ME , the interacting dynamics
between S and an element E of the chain C. The explicit form of the operator V is
dictated by the underlying physics, we give some examples in Section 3.
For m ≥ 1 let us denote by
X
e m = Lm +
L
LE,k
(2.11)
k6=m
BJM November 4, 2005
7
the generator of the total dynamics during the interval [(m − 1)τ, mτ ). We have introduced Lm , the operator on H that acts trivially on all elements of the chain except for
the m–th one. On the remaining part of the space (which is isomorphic to HS ⊗ HE ),
Lm acts as L, (2.10). We have also set LE,k to be the operator on H that acts nontrivially only on the k–th element of the chain, on which it equals LE . Of course, the
infinite sum in (2.11) must be interpreted in the strong sense on H.
Decompose t ∈ R+ as
t = m(t)τ + s(t),
(2.12)
where m(t) is the integer measuring the number of complete interactions of duration
τ the system S has undergone at time t, and where 0 ≤ s(t) < τ . The repeated
interaction dynamics of an operator A on H is defined by
t
αRI
(A) = URI (t)∗ A URI (t)
(2.13)
where
URI (t) = e−is(t)Lm(t)+1 e−iτ Lm(t) · · · e−iτ L1
e
e
e
(2.14)
defines the Schrödinger dynamics on H. According to this dynamics S interacts in
succession, for a fixed duration τ and a fixed interaction V , with the first m(t) elements
of the chain, and for the remaining duration s(t) with the (m(t) + 1)–th element of the
chain. Being the propagator of a “time-dependent Hamiltonian” (which is piecewise
constant), URI (t) does not have the group property in t.
Our goal is to examine the large time behaviour of expectation values of certain
observables in normal states ω on M (states given by a density matrix on H). The
system S feels an effective dynamics induced by the interaction with the chain C. Under
a suitable ergodicity assumption on this effective dynamics the small system is driven
to an asymptotic state, as time increases. We will express the effective dynamics and
the ergodic assumption using the modular data of the pair (MS ⊗ MC , ΩS ⊗ ΩC ).
Let J and ∆ denote the modular conjugation and the modular operator associated
to (MS ⊗ ME , ΩS ⊗ ΩE ), [5]. We assume that
(A) ∆1/2 V ∆−1/2 ∈ MS ⊗ ME
and we introduce the operator
K = L − J∆1/2 V ∆−1/2 J,
(2.15)
called a C–Liouville operator, [6, 9]. It generates a strongly continuous group of
1/2
−1/2 k
bounded operators, denoted eitK , satisfying keitK k ≤ e|t| k∆ V ∆
. The main feature of the operator K is that eitK implements the same dynamics as eitL on MS ⊗ ME
(since the difference K − L belongs to the commutant M0S ⊗ M0E ), and that
KΩS ⊗ ΩE = 0.
(2.16)
Relation (2.16) follows from assumption (A), definition (2.15) and the properties ∆−1/2 J =
J∆1/2 and J∆1/2 AΩS ⊗ ΩE = A∗ ΩS ⊗ ΩE , for any A ∈ MS ⊗ ME .
BJM November 4, 2005
8
Let
P = 1lHS ⊗ |ΩC ihΩC |
(2.17)
be the orthogonal projection onto HS ⊗ CΩC ∼
= HS , where ΩC is given in (2.2). If B is
an operator acting on H then we identify P BP as an operator acting on HS . We have
Proposition 2.1 There is a constant C < ∞ s.t. k(P eitK P )m kB(HS ) ≤ C, for all
t ∈ R, m ≥ 0. In particular, spec(P eitK P ) ⊂ {z ∈ C | |z| ≤ 1} and all eigenvalues
lying on the unit circle are semisimple.
We give a proof of Proposition 2.1 in Section 4.3. Relation (2.16) implies that for all
t ∈ R, P eitK P ΩS = ΩS . Our assumption (E) on the effectiveness of the coupling is an
ergodicity assumption on the discrete dynamics generated by
M ≡ M (τ ) = P eiτ K P
(2.18)
(E) The spectrum of M on the complex unit circle consists of the single eigenvalue
{1}. This eigenvalue is simple (with corresponding eigenvector ΩS ).
Assumption (E) guarantees that the adjoint operator M ∗ has a unique invariant vector,
called Ω∗S (normalized as hΩ∗S , ΩS i = 1), and that
lim M m = π := |ΩS ihΩ∗S |,
m→∞
(2.19)
in the operator sense, where π is the rank one projection which projects onto CΩS
along (CΩ∗S )⊥ . In fact, we have the following easy estimate (valid for any matrix M
with spectrum inside the unit disk and satisfying (E))
Proposition 2.2 For any > 0 there exists a constant C s.t. kM m −πk ≤ C e−m(γ−) ,
for all m ≥ 0, where γ := minz∈spec(M )\{1} | log |z| | > 0.
The parameter γ measures the speed of convergence.
Remark. If all eigenvalues of M are semisimple then in Proposition 2.2 we have
kM m − πk ≤ Ce−mγ for some constant C and all m ≥ 0.
As a last preparation towards an understanding of our results we discuss the kinds
of observables we consider. One interesting such class is MS ⊂ M which consists of
observables of the system S only. There are other observables of interest. We may
think of the system S as being fixed in space and of the chain as passing by S so that
at the moment t, the (m(t) + 1)–th element E is located near S, c.f. (2.12). A detector
placed in the vicinity of S can measure at this moment in time observables of S and
those of the (m(t) + 1)–th element in the chain, i.e., an “instantaneous observable” of
the form AS ⊗ ϑm(t)+1 (B0 ), where AS ∈ MS , B0 ∈ ME , and ϑm : ME → MC is defined
by
ϑm (AE ) = 1lE · · · 1lE ⊗ AE ⊗ 1lE · · ·
(2.20)
where the AE on the right side of (2.20) acts on the m–th factor in the chain. An
example of such an observable is the energy flux (variation) of the system S. More
generally we may be interested in the expectation value of operators of the form
[AS ; Ai ; Bj ] ≡ AS ⊗pi=1 Ai ⊗rj=−` ϑm(t)+j+1 (Bj ),
(2.21)
BJM November 4, 2005
9
where AS ∈ MS , A1 , . . . , Ap ∈ ME , B−` , . . . , B0 , . . . , Br ∈ ME and where t = m(t)τ +
s(t) as in (2.12) and ϑk is given in (2.20). The parameters p ≥ 1, `, r ≥ 0 are not
displayed in the l.h.s. in (2.21). (We always assume that p < m(t) − ` + 1.) AS and
the Ai represent observables we measure on the small system and on the element with
index i of the chain, the B0 is the “instantaneous” observable, measured in the element
m(t) + 1 of the chain (the one in contact with S at time t), while the Bj with negative
and positive index are the quantities measured in the elements preceding and following
the (m(t) + 1)–th.
2.2
Asymptotic state
Throughout the paper we assume that Conditions (A) and (E) of the previous section
are satisfied.
We consider the large time limit of expectations
t
E(t) = ω αRI
[AS ; Ai ; Bj ]
(2.22)
for observables [AS ; Ai ; Bj ] as in (2.21) and for normal states ω on M. Define the state
ω+ on MS by
ω+ (AS ) = hΩ∗S , AS ΩS i ,
(2.23)
where Ω∗S is defined in (2.19).
Theorem 2.3 Let ω be fixed and take Ai = 1lE , i = 1, . . . , p. For any > 0 there is a
constant C s.t. for all t ≥ 0
|E(t) − E+ (t)| ≤ C e−t(γ−)/τ ,
(2.24)
where γ > 0 is given in Proposition 2.2, and where E+ is the τ -periodic function
Q
`τ +s(t)
r
E+ (t) = ω+ P αRI
AS ⊗ B−` ⊗ · · · ⊗ B0 P
(2.25)
j=1 hBj iΩE .
Here, hBj iΩE = hΩE , Bj ΩE i.
We define the RIAS to be the τ –periodic state on MS ⊗rj=−` ME given by
Q
`τ +s(t)
r
AS ⊗rj=−` Bj 7→ ω+ P αRI
AS ⊗ B−` ⊗ · · · ⊗ B0 P
j=1 hBj iΩE .
(2.26)
Using (2.24) and the unicity of the limit, one can see that actually the state ω+ does
not depend on the choice of the reference state ΩS .
Remarks. 1) If B−` , . . . , B−`0 −1 = 1lE for some −`0 − 1 ≤ −1, then one shows that
`τ +s(t)
`0 τ +s(t)
ω+ P αRI
AS ⊗ B−` ⊗ · · · ⊗ B0 P = ω+ P αRI
AS ⊗ B−`0 ⊗ · · · ⊗ B0 P ,
`τ +s(t)
and in case Bj = 1lE for all j = −`, . . . , 0 formula (2.25) is understood with αRI
s(t)
replaced by αRI .
BJM November 4, 2005
10
2) C inTheorem 2.3 is uniform in τ for τQ> 0 varying in compact
sets, and it is
uniform in AS ∈ MS , {Bj }rj=1 ⊂ ME kAS k rj=1 kBj k ≤ const. .
3) The convergence is determined by that of Proposition 2.2. If the ergodic assumption (E) is not satisfied then the limit limn→∞ M n still exists, in a weaker sense.
Namely, if there are eigenvalues
from 1 on the circle, then the limit exists
P different
−1
n = π + O( 1 ). Further, if 1 is a degenerate
in the ergodic mean sense, N1 N
M
n=0
N
eigenvalue of M then the limit exists but the projection π is not one dimensional. This
reflects in Theorem 2.3 in the following way. If 1 is non degenerate, but there are other
eigenvalues on the circle, then Theorem 2.3 holds with (2.24) replaced by
m(t)
1 X
C
E
(t)
+
≤ .
E(m(t)τ + s(t)) −
(2.27)
t
τ t
m=0
If on the other hand 1 is degenerate but there is no other eigenvalue on the circle,
then one can still prove that the expectation value E(t) has an aymptotic behaviour
E∞ (t, ω), which is τ −periodic, but which will a priori depend on the initial state ω
(c.f. (4.25) in the proof of Theorem 2.3). Of course if both 1 is degenerate and there
are other eigenvalues on the circle, then one gets convergence to E∞ (t, ω) but in the
ergodic mean.
Our next result incorporates the measurement of observables A1 , . . . Ap ∈ ME for
a chain consisting of dispersive systems E. We measure dispersivity by the property of
return to equilibrium. E is said to have the latter property iff for any normal state ωE
on ME we have the relation
lim ωE (τEt (AE )) = hΩE , AE ΩE i ,
t→∞
(2.28)
for any AE ∈ ME . Examples of such E include reservoirs of ideal quantum gases. It
is worthwile to mention that E has the property of return to equilibrium if and only if
eitLE converges in the weak sense to the orthogal projection onto CΩE , as t → ∞.
Theorem 2.4 Suppose E has the property of return to equilibrium. Then
lim |E(t) − E+ (t)| = 0,
t→∞
where E+ (t) is the τ -periodic function
Q
Qr
`τ +s(t)
p
E+ (t) = ω+ P αRI
AS ⊗ B−` ⊗ · · · ⊗ B0 P
j=1 hBj iΩE .
i=1 hAi iΩE
(2.29)
(2.30)
Remark. The speed of convergence in (2.29) is determined by that of return to equilibrium, (2.28), and by γ, Proposition 2.2. The limit (2.29) is uniform
in τQ
, for τ varying
Q
in compact sets, and it is uniform in balls of observables kAS k pi=1 kAi k rj=−` kBj k ≤
const.
BJM November 4, 2005
2.3
11
Correlations & reconstruction of initial state
As Theorems 2.3 and 2.4 show, the limiting expectation values E+ (t) are independent
of the initial state (the state ω+ is, c.f. (2.23)). However, limiting correlations are not,
and their knowledge allows to reconstruct the initial state.
Fix a normal initial state ω of M and let A ∈ M, AS ∈ MS , B0 ∈ ME . We define
the correlation between A and the instantaneous observable AS ⊗ ϑm(t)+1 (B0 ) by
t
C(t; A, AS , B0 ) = ω A αRI
(AS ⊗ ϑm(t)+1 (B0 )) .
(2.31)
Theorem 2.5 For any > 0 there is a constant C s.t. for all t ≥ 0
| C(t; A, AS , B0 ) − C+ (t; A, AS , B0 )| ≤ C e−t(γ−)/τ ,
(2.32)
where γ is given in Proposition 2.2, and where C+ is the τ –periodic limiting correlation
function
s(t)
C+ (t; A, AS , B0 ) = ω(A) ω+ P αRI (AS ⊗ B0 )P ,
(2.33)
with ω+ defined in (2.19).
Remark. Relation (2.33) allows us to reconstruct the initial state ω, knowing the
asymptotic state ω+ and the asymptotic correlation function C+ .
2.4
Energy, entropy and their relation
It may not be meaningful to speak about the total energy of the system, because it
may have to be considered as being infinite, e.g. if the elements E of the chain are
infinitely extended quantum systems with non-vanishing energy density. However, we
can define the time variation of the total energy of the system and link it to its entropy
variation, giving us an average 2nd law of thermodynamics for RIAS.
2.4.1
Energy
e m+1 , m ≥ 0, is the generator of the total dynamics in the time interval
Recall that L
t = mτ +s ∈ [mτ, (m+1)τ ), during which the (m+1)–th element of the chain interacts
with S, c.f. (2.11). Given any integer m ≥ 0 and any 0 ≤ s < τ it is easy to formally
verify the relation
mτ +s e
mτ e
(Lm+1 ) − αRI
(Lm+1 ) = 0.
(2.34)
αRI
t (L
e m+1 ) is constant for t in any interval
This suggests that the formal quantity αRI
[mτ, (m + 1)τ ). Another short calculation yields that this quantity undergoes a jump
j(k) as time passes the moment kτ , k ≥ 1: for (k − 1)τ ≤ t1 < kτ ≤ t2 < (k + 1)τ we
have
t2 e
t1 e
kτ
j(k) := αRI
(Lk+1 ) − αRI
(Lk ) = αRI
Vk+1 − Vk ,
(2.35)
where we set
Vk = [1lMS ⊗ ϑk ](V )
(2.36)
(see (2.20)). We interpret j(k) as the change in total energy as time passes the moment
kτ .
BJM November 4, 2005
12
Theorem 2.3 tells us that for any normal state ω on M and for any > 0, there is
a constant C s.t.
ω(j(k)) − ω+ j+ ≤ C e−k(γ−) ,
(2.37)
where
j+ = P V P −
τ
P αRI
(V
Z
)P = −i
τ
s
P αRI
[LS + LE , V ] P ds.
(2.38)
0
Relation (2.37) and the fact that the energy is piecewise constant shows that ω+ (j+ )
is the change of energy in any interval of length τ , in the large time limit. We thus call
dE+ =
1
ω+ (j+ )
τ
(2.39)
the asymptotic energy production. The asymptotic energy production does not depend
on the initial state of the system.
Remark. It is not hard to see that the expectation of the energy jump is constant
in the state ω+ ⊗ ωC , where ωC is the vector state on MC determined by ΩC , (2.2):
ω+ ⊗ ωC (j(k)) = ω+ (j+ ),
∀k ≥ 1.
(2.40)
We introduce the variation of the total energy, ∆E(t), between the instants t =
m(t)τ + s(t) and t = 0. It is the sum of the energy jumps,
m(t)
∆E(t) =
X
j(k), for t ≥ τ ,
(2.41)
k=1
and ∆E(t) = 0 if 0 ≤ t < τ . Estimate (2.37) shows that for any normal state ω on M
there is a constant C s.t.
ω(∆E(t))
C
− dE+ ≤ ,
(2.42)
t
t
for all t > 0. The energy grows asymptotically linearly in time.
2.4.2
Entropy, average 2nd law of thermodynamics
Let ω and ω0 be two normal states on M. The relative entropy of ω with respect to ω0
is denoted by Ent(ω|ω0 ), where our definition of relative entropy differs from that one
given in [5] by a sign, so that in our case, Ent(ω|ω0 ) ≥ 0.
For a thermodynamic interpretation of the entropy and its relation to the energy
variation, we assume in this section that ΩS is a (βS , τSt )–KMS state on MS , and that
ΩE is a (βE , τEt )–KMS state on ME , where βS , βE are inverse temperatures. Let ω0 be
the state on M determined by the vector ΩS ⊗ ΩC (c.f. before (2.1), and (2.2)).
We are interested in the change of relative entropy of the repeated interaction
system as time evolves.
t |ω
Proposition 2.6 Let ω be any normal state on M. Then Ent ω ◦ αRI
0 is a continuous, piecewise differentiable function of t ≥ 0. Moreover, we have
t
t
Ent ω ◦ αRI
|ω0 − Ent(ω|ω0 ) = ω βE ∆E(t) − αRI
(X(t)) + X(0) ,
(2.43)
BJM November 4, 2005
13
where ∆E(t) is the variation of the total energy between the moments t = 0 and t =
m(t)τ + s(t), (2.41), and where
X(t) = βE Vm(t)+1 + (βE − βS )LS ,
(2.44)
with Vk given by (2.36).
The proof of (2.43) is based on the entropy production formula [7]. We give it in Section
4.4. It is not hard to verify that for t ∈ (mτ, (m + 1)τ ) we have
d
t
t
Ent ω ◦ αRI
|ω0 = −ω αRI
i[βS LS + βE LE,m(t)+1 , Vm(t)+1 ] ,
(2.45)
dt
t |ω
and that left and right derivatives of Ent ω ◦ αRI
0 exist as t → mτ , but they do not
coincide.
If Ent(ω|ω0 ) < ∞ then all terms in (2.43) are bounded uniformly in t, except
t |ω ) and ω(β ∆E(t)). Hence (2.39) and (2.42) show that for any
possibly Ent(ω ◦ αRI
0
E
normal state ω on M there is a constant C s.t.
t |ω )
Ent(ω ◦ αRI
C
βE
0
≤ ,
−
ω
(j
)
(2.46)
+
+
t
τ
t
for all t > 0. The entropy grows linearly in time, for large times.
The relative entropy is non–negative, so (2.46) shows that
ω+ (j+ ) ≥ 0.
(2.47)
We show in Section 3 that ω+ (j+ ) is strictly positive for concrete systems. It follows
from (2.46) also that
t
(2.48)
|ω0 ) < ∞ ⇐⇒ ω+ (j+ ) = 0.
sup Ent(ω ◦ αRI
t≥0
Since ω+ (j+ ) is independent of ω it follows that for a given interaction (V, τ ) the relative
entropy either diverges for all initial states ω, as t → ∞, or it stays bounded for all
initial states ω. In particular, if ω+ (j+ ) > 0 then there does not exist any normal state
t (i.e., such that ω ◦ αt = ω, for all t ≥ 0).
ω on M which is invariant under αRI
RI
Proposition 2.7 We have
t+τ
t
|ω0 ) − Ent(ω ◦ αRI
|ω0 ) = βE ω+ (j+ ).
lim Ent(ω ◦ αRI
t→∞
(2.49)
The change of entropy during an interval of duration τ , for t → ∞, is thus given
by βE ω+ (j+ ) ≥ 0. We call
βE
dS+ =
ω+ (j+ )
(2.50)
τ
the (average) asymptotic entropy production. The quantity dS+ represents the increase
in entropy per unit time, in the limit of large times. It does not depend on the initial
state of the system.
Remark. One sees easily that the expectation of dS+ is constant in the state
ω+ ⊗ ωC (see also (2.40)).
Relations (2.50) and (2.39) lead us to the average 2nd law of thermodynamics,
dE+ = TE dS+ ,
TE = 1/βE .
This law does not depend on the initial state of the system.
(2.51)
BJM November 4, 2005
3
3.1
14
Examples
Spin-Fermion system with quadratic interaction
As our main example, we consider the case where the small system S is a 2−level system
and the elements of the chain consist of free Fermi reservoirs at positive temperature
β −1 . Let us first describe precisely the model (see also [6] and references therein).
The von Neumann algebra of observables for the small system is
MS = M2 (C) ⊗ 1l = {A ⊗ 1l|A ∈ M2 (C)}
(3.1)
acting on the Hilbert space
H S = C2 ⊗ C 2 .
0 1
Let σx , σy , σz be the usual Pauli matrices, i.e. σx =
, σy =
1 0
1 0
σz =
. The dynamics of the small system is then given by
0 −1
τSt (A ⊗ 1l) = eitσz Ae−itσz ⊗ 1l.
(3.2)
0 i
,
−i 0
(3.3)
For convenience we chose the reference state ωS to be the tracial state, i.e. ωS (A⊗1l) =
1
t
2 Tr(A). Note that it is a (τS , 0)−KMS state. Its representative vector is
1
1
ΩS = √ ψ 1 ⊗ ψ 1 + √ ψ 2 ⊗ ψ 2
2
2
(3.4)
where (ψ1 , ψ2 ) is the canonical basis of C2 . For shortness, we will denote by ψij :=
ψi ⊗ ψj the corresponding basis of HS . The standard Liouvillean then writes
LS = σz ⊗ 1l − 1l ⊗ σz ,
(3.5)
and its spectrum is spec(LS ) = {−2, 0, 2} where 0 has multiplicity 2, and −2, 2 are
non degenerate. Finally, the modular conjugation and modular operator associated to
(MS , ΩS ) are
JS (φ ⊗ ψ) = ψ̄ ⊗ φ̄, ∆S = 1l ⊗ 1l,
(3.6)
and where ¯· denotes the usual complex conjugation on C2 .
We then describe an element of the chain, i.e. a free Fermi gas at inverse temperature β. Let h be the Hilbert space of one single fermion and h its energy operator.
The operators a(f ) and a∗ (f ) denote the corresponding annihilation and creation operators acting on the fermionic Fock space Γ− (h) and they satisfy the canonical anticommutation relations (CAR). As a consequence of the CAR, the operators a(f ) and
a∗ (f ) are bounded and satisfy ka# (f )k = kf k where a# stands either for a or for
a∗ . The algebra of observables of a free Fermi gas is the C ∗ -algebra of operators A
generated by {a# (f )|f ∈ h}. The dynamics is then given by
τft (a# (f )) = a# (eith f ).
(3.7)
BJM November 4, 2005
15
It is well known (see e.g. [5, 10]) that for any β > 0, there is a unique (τf , β)−KMS
state ωβ on A which is determined by the two point function ωβ (a∗ (f )a(f )) = hf, (1 +
eβh )−1 f i. Finally, let Ωf be the Fock vacuum and N the number operator.
We now fix a complex conjugation (anti-unitary involution) f → f¯ on h which
commutes with the energy operator h. It naturally extends to a complex conjugation
on the Fock space Γ− (h) and we denote it by the same symbol, i.e. Φ → Φ̄.
The GNS representation of the algebra A associated to the KMS-state ωβ is the
triple (HF , πβ , ΩF ) [1] where
HF = Γ− (h) ⊗ Γ− (h),
ΩF = Ωf ⊗ Ωf ,
(3.8)
and
eβh/2
1
N
∗
¯
√
√
πβ (a(f )) = a
f ⊗ 1l + (−1) ⊗ a
f =: aβ (f ),
βh
βh
1+e
1+e
βh/2
e
1
∗
∗
N
¯
πβ (a (f )) = a √ βh f ⊗ 1l + (−1) ⊗ a √ βh f =: a∗β (f ).
1+e
(3.9)
1+e
The von Neumann algebra of observables for an element of the chain will then be
the enveloping von Neumann algebra
ME = πβ (A)00 ,
(3.10)
HE = HF .
(3.11)
τEt (πβ (A)) = πβ (τft (A))
(3.12)
acting on the Hilbert space
The dynamics on πβ (A) is given by
and extends to ME in a unique way. The representative vector of the equilibrium state
is
ΩE = ΩF ,
(3.13)
and the standard Liouvillean then writes
LE = dΓ(h) ⊗ 1l − 1l ⊗ dΓ(h).
(3.14)
Finally the modular conjugation and the modular operator associated to (ME , ΩE ) are
JE (Φ ⊗ Ψ) = (−1)N (N −1)/2 Ψ̄ ⊗ (−1)N (N −1)/2 Φ̄,
∆E = e−βLE .
(3.15)
We finally specify the interaction between the small system and the elements of the
chain, i.e. the operator V . Let g ∈ h be a form factor, we set
V := σx ⊗ 1lC2 ⊗ a∗β (g)aβ (g) ∈ MS ⊗ ME .
(3.16)
This is the simplest non trivial interaction for which the number of particles is conserved.
We moreover assume that
BJM November 4, 2005
(SF1)
16
eβh/2 g ∈ h.
This ensures that Assumption (A) is satisfied. Indeed, using (3.6)-(3.15)-(3.16), we get
∆1/2 V ∆−1/2
"
= σx ⊗ 1lC 2 ⊗
(3.17)
a∗
×
√
1
g ⊗ 1l + (−1)N ⊗ a
e−βh/2
!!
√
ḡ
1+
1 + eβh
!!#
eβh
eβh/2
N
∗
√
a √
g ⊗ 1l + (−1) ⊗ a
ḡ
.
1 + eβh
1 + eβh
eβh
The Liouville operator which generates the interacting dynamics is then the selfadjoint
operator
Lλ := LS + LE + λV,
(3.18)
while the C–Liouville operator is
Kλ := LS + LE + λ(V − J∆1/2 V ∆−1/2 J) = K0 + λW,
(3.19)
and where λ ∈ R is a coupling constant.
For simplicity reasons we will moreover assume that
(SF2) h = L2 (R+ , g) where g is some auxiliary Hilbert space and the operator h is the
multiplication operator by r ∈ R+ .
Finally, let gβ (r) := (1 + e−βr )−1/2 g(r).
Note also that the system E has the property of return to equilibrium.
Our first result is the
Theorem 3.1 Suppose Assumption (SF1)-(SF2) are satisfied. Then for any τ ∈
/ π2 +
πN, there exists Λ0 > 0 such that for all 0 < |λ| < Λ0 , the operator Mλ := P eiτ Kλ P satisfies the ergodic assumption (E). In particular the spin-fermion system with quadratic
2
interaction satisfies Theorem 2.3, with γ = τ (α12+α2 ) λ2 + O(λ3 ), Theorem 2.4, and
moreover the asymptotic state ω+,λ is given by
ω+,λ (AS ) =
1
hα1 ψ11 + α2 ψ22 , AS (ψ11 + ψ22 )i
(3.20)
α1 + α2
2 α − α
1
2
hψ12 + ψ21 , AS (ψ11 + ψ12 )i + O(λ2 ),
+λ e−βh/2 gβ h 2(α1 + α2 )
where, for j = 1, 2,
Z Z
−βrj
2
2
2 τ (2 − r1 + r2 )
αj :=
dr1 dr2 e
kgβ (r1 )kg kgβ (r2 )kg sinc
> 0,
2
sinc(x) :=
sin x
x ,
and all the integrals run over R+ .
(3.21)
BJM November 4, 2005
17
The unperturbed operator M0 has eigenvalues 1 (with multiplicity 2), e2iτ and e−2iτ
(see (3.5)). The assumption on the interaction time τ ensures that these eigenvalues do
not coincide and makes the computation in perturbation theory as simple as possible.
However, it can probably be omitted.
One can also see that the asymptotic expectation E+ (t) (see (2.25)) has a non
trivial periodicity (i.e. it is not constant) at the order λ.
We now turn to the question of entropy production for this simple model. We will
prove that it is strictly positive, at least for small coupling constant. More precisely,
we have
Theorem 3.2 Suppose Assumptions (SF1)-(SF2) are satisfied. Then for any τ ∈
/
π
2 + πN, there exists Λ1 > 0 such that, for all 0 < |λ| < Λ1 , the system has strictly
positive asymptotic entropy production.
3.2
Spin-Fermion system with linear interaction
As a second example, we consider the same spin fermion system at inverse temperature
β. The only change concerns the interaction term. Let g ∈ h be a form factor, we set
V := σx ⊗ 1lC2 ⊗ (aβ (g) + a∗β (g)).
(3.22)
Once again, we assume that Assumption (SF1) holds which ensures that ∆1/2 V ∆−1/2 ∈
M. Indeed
1
eβh
∗
1/2
−1/2
∆ V∆
= σx ⊗ 1lC 2 ⊗ a √
g ⊗ 1l + a √
g ⊗ 1l
1 + eβh
1 + eβh
!
!#
eβh/2
e−βh/2
N
∗
N
√
+(−1) ⊗ a
ḡ + (−1) ⊗ a √
ḡ
.
1 + eβh
1 + eβh
We then have the same kind of result as for the case of quadratic interaction, namely
Theorem 3.3 Suppose Assumption (SF1)-(SF2) are satisfied. Then for any τ ∈
/
π
iτ Kλ P
+
πN,
there
exists
Λ
>
0
such
that
for
all
0
<
|λ|
<
Λ
,
the
operator
M
:=
P
e
0
0
λ
2
satisfies the ergodic assumption (E). In particular the spin-fermion system with linear
2
interaction satisfies Theorem 2.3, with γ = τ (α12+α2 ) λ2 + O(λ3 ), Theorem 2.4 and the
asymptotic state ω+,λ is given by
ω+,λ (AS ) =
1
hα1 ψ11 + α2 ψ22 , AS (ψ11 + ψ22 )i + O(λ2 ),
α1 + α2
(3.23)
where
τ (r − 2)
2 τ (r + 2)
:=
e sinc
+ sinc
,
2
2
Z
τ (r + 2)
τ (r − 2)
:=
dr kgβ (r)k2g e−βr sinc2
+ sinc2
.
2
2
Z
α1
α2
dr kgβ (r)k2g
−βr
2
and all the integrals run over R+ . Moreover the system has strictly positive asymptotic
entropy production.
BJM November 4, 2005
3.3
18
Spin-Spin model
As our last example, we consider a model in which the small system as well as the
elements of the chain consist of a 2−level system. Such kind of systems (or more
generally a d−level system interacting with a chain of n−level systems) have been
considered previously in [2].
The von Neumann algebra of observables for the small system and for the elements
of the chain is
MS = ME = M2 (C) ⊗ 1l = {A ⊗ 1l|A ∈ M2 (C)}
(3.24)
acting on the Hilbert space
H S = H E = C2 ⊗ C2 .
(3.25)
Let ES and EE be non negative real numbers. They will play the role of the energy of
the “excited” state of the small system and of the elements of the chain respectively.
The dynamics of the small system is then given by
τSt (A ⊗ 1l) = eithS Ae−ithS ⊗ 1l,
(3.26)
and the one of an element of the chain by
τEt (A ⊗ 1l) = eithE Ae−ithE ⊗ 1l,
(3.27)
0 0
0 0
where hS =
and hE =
.
0 ES
0 EE
Once again, for convenience, we chose the reference state ωS to be the tracial state,
i.e. ωS (A ⊗ 1l) = 21 Tr(A) (the results of course do not depend on this choice). Its
representative vector is (Section 3.1)
1
1
ΩS = √ ψ11 + √ ψ22 .
2
2
(3.28)
The standard Liouvillean writes
LS = hS ⊗ 1l − 1l ⊗ hS ,
(3.29)
and the modular conjugation and modular operator associated to (MS , ΩS ) are
JS (φ ⊗ ψ) = ψ̄ ⊗ φ̄,
∆S = 1l ⊗ 1l.
(3.30)
In order to avoid confusions between the small system and an element of the chain
we will denote by φij = φi ⊗ φj instead of ψij the basis of HE . The reference state ωE
will be the (τE , β)−KMS state. Its representative vector writes
ΩE = √
1
1 + e−βEE
(φ11 + e−βEE /2 φ22 ).
(3.31)
The standard Liouville operator is
LE = hE ⊗ 1l − 1l ⊗ hE ,
(3.32)
BJM November 4, 2005
19
and the modular conjugation and modular operator associated to (ME , ΩE ) are
JE (φ ⊗ ψ) = ψ̄ ⊗ φ̄,
∆E = e−βLE .
(3.33)
Note that here the system E does not have the property of return to equilibrium.
We now describe the interaction. Let us denote by a and a∗ the annihilation and
creation operators associated to the vectors φ1 (ground state)
and φ2(excited state),
a b
i.e. aφ1 = 0, aφ2 = φ1 , a∗ φ1 = φ2 , a∗ φ2 = 0. Finally let I =
∈ M2 (C). The
c d
interaction is then given by
V = I ⊗ 1l ⊗ a∗ ⊗ 1l + I ∗ ⊗ 1l ⊗ a ⊗ 1l.
(3.34)
The Liouville operator which generates the interacting dynamics is then the selfadjoint
operator
Lλ := LS + LE + λV,
(3.35)
while the C–Liouville operator is
Kλ := LS + LE + λ(V − J∆1/2 V ∆−1/2 J) = K0 + λW,
(3.36)
where λ is a coupling constant.
We finally consider the following assumptions.
(SS1) b 6= 0 and τ (EE − ES ) ∈
/ 2πZ.
(SS2) c 6= 0 and τ (EE + ES ) ∈
/ 2πZ.
If either (SS1) or (SS2) is satisfied, then the ergodic assumption (E) holds.
Theorem 3.4 Suppose that τ EE ∈
/ πZ and that either Assumption (SS1) or (SS2)
is satisfied. Then, there exists Λ0 > 0 such that for all 0 < |λ| < Λ0 , the operator
Mλ := P eiτ Kλ P satisfies the ergodic assumption (E). In particular the spin-spin system
satisfy Theorem 2.3, with γ = γ0 λ2 + O(λ3 ), and the asymptotic state ω+,λ is given by
ω+,λ (AS ) =
1
hα1 ψ11 + α2 ψ22 , AS (ψ11 + ψ22 )i + O(λ2 ),
α1 + α2
(3.37)
where
α1
α2
γ0
τ (EE − ES )
−βEE
2
2 τ (EE + ES )
:= |b| sinc
+e
|c| sinc
≥ 0,
2
2
τ (EE − ES )
τ (EE + ES )
:= e−βEE |b|2 sinc2
+ |c|2 sinc2
≥ 0,
2
2
2
τ (α1 + α2 ) τ 2 (α1 + α2 )
:= min
,
1 + e−βEE 2(1 + e−βEE )
τ2
2 τ EE
2
2
+ sinc (
)(|a| + |d| − ād − ad̄) .
2
2
2
2
If moreover both Assumptions (SS1) and (SS2) are satisfied, then the system has
strictly positive asymptotic entropy production.
BJM November 4, 2005
20
Once again, we assume that τ EE ∈
/ πZ just in order to make the eigenvalues of
M0 not coincide and this can probably be weakened. On the other hand, Assumptions
(SS1) and (SS2) are much deeper. Their signification is that there is an effective
coupling between the ground state and the excited state of the small system (b or c non
zero) as well as a non resonant phenomenon between the energies of the small system
and the elements of the chain (τ (EE − ES ) or τ (EE + ES ) not in 2πZ). Asking that
either (SS1) or (SS2) is satisfied is actually equivalent to the condition α1 + α2 6= 0.
4
4.1
Proofs
Proofs of Theorems 2.3, 2.4
We give the full proof of Theorem 2.4, the proof of Theorem 2.3 is a special case of the
former.
It is enough to show (2.29), (2.30) for vector states ω(·) = hψ, · ψi, ψ ∈ H, kψk =
1. Further, since every ψ ∈ H is approximated in the norm of H by finite linear
combinations of vectors of the form ψS ⊗m≥1 ψm , where ψS ∈ HS , ψm = ΩE if m > N ,
for some N < ∞, it suffices to prove (2.29), (2.30) for vector states determined by
vectors of the form
ψS ⊗N
(4.1)
m=1 ψm ⊗m>N ΩE ∈ H,
where kψS k = kψm k = 1, 1 ≤ m ≤ N , for arbitrary N < ∞. Finally, since the
vectors ΩS , ΩE are cyclic for the commutants M0S , M0E , any vector of the form (4.1) is
approximated by a
ψ = B 0 ΩS ⊗ ΩC ,
(4.2)
for some
0
0
B 0 = BS0 ⊗N
m=1 Bm ⊗m>N 1lE ∈ M ,
(4.3)
0 ∈ M0 . It is therefore sufficient to show (2.29), (2.30) for vectors
with BS0 ∈ M0S , Bm
E
of the form (4.2), (4.3).
Let AS ∈ MS , A1 , . . . , Ap ∈ ME and B−` , . . . , B0 , . . . , Br ∈ ME be fixed observables
(`, r ≥ 0). We examine the expectation value
t
(4.4)
E(t) := ψ, αRI
AS ⊗pi=1 Ai ⊗rj=−` ϑm(t)+j+1 (Bj ) ψ ,
where ψ is given in by (4.2), m(t) is determined by (2.12), and where the Ai act on
the first p factors of HC . It is clear that the trivial factors 1lE are omitted in (4.4). The
choice of indices in (4.4) is such that at time t the observable B0 is measured in the
element E of the chain which is interacting with S (i.e., the (m(t) + 1)–th element of
the chain). The following decomposition serves to isolate the dynamics of the elements
E which do not interact at a given time, see also (2.11).
− −isLm+1 −iτ Lm
+
e−isLm+1 e−iτ Lm · · · e−iτ L1 = Um
e
e
· · · e−iτ L1 Um
,
e
e
e
(4.5)
BJM November 4, 2005
21
where

−
Um
= exp −i
m
X

[(m − j)τ + s]LE,j  ,
(4.6)
j=1

+
Um
= exp −i
m+1
X

(j − 1)τ LE,j − i(mτ + s)
j=2
X
LE,j  .
(4.7)
j≥m+2
We obtain from (2.13), (2.14) and (4.5)
+ ∗ iτ L1
E(t) = ψ, (Um
) e
· · · eiτ Lm eisLm+1
(m−i)τ +s
(−j−1)τ +s
×AS ⊗pi=1 τE
(Ai ) ⊗−1
j=−` ϑm+j+1 (τE
−isLm+1 −iτ Lm
−iτ L1 +
×e
e
···e
Um ψ ,
(Bj )) ⊗rj=0 ϑm+j+1 (Bj )
(4.8)
+ )∗ · · · U + in the r.h.s. of (4.8)
where we write m, s for m(t), s(t). The operator (Um
m
belongs to M. Hence we can commute the B 0 in (4.2) with this operator to obtain
+ ∗ iτ L1
· · · eiτ Lm eisLm+1
E(t) = Ω, (B 0 )∗ B 0 (Um
) e
(m−i)τ +s
×AS ⊗pi=1 τE
(−j−1)τ +s
(Ai ) ⊗−1
j=−` ϑm+j+1 (τE
×e−isLm+1 e−iτ Lm · · · e−iτ L1 Ω
r
Y
(Bj )) ⊗ ϑm+1 (B0 )
hBj iΩE ,
(4.9)
j=1
where we have set
Ω = ΩS ⊗ ΩC ,
(4.10)
+
and we write hOiχ = hχ, Oχi for an operator O and a vector χ. The operator Um
disappears because it leaves Ω invariant, c.f. (2.9). We are able to factorize the averages
hBj iΩE , for j ≥ 1, because B 0 and all propagators in (4.8) act trivially on factors of HC
with index ≥ m + 2 (note also that N < m since we have in mind the limit m → ∞).
Since (B 0 )∗ B 0 acts trivially on factors in HC with index ≥ N + 1 we may replace
+
+
(Um )∗ = (Um(t)
)∗ in (4.9) by

eN = exp i
U
N
X

(j − 1)τ LE,j  ,
(4.11)
j=1
which is a unitary not depending on t. Without changing the value of (4.9) we can
replace in that equation successively L1 by K1 , then L2 by K2 , up to replacing Lp by
Kp . Then
eN eiτ K1 · · · eiτ Kp eiτ Lp+1 · · · eiτ Lm eisLm+1
E(t) = Ω, (B 0 )∗ B 0 U
(−j−1)τ +s
×AS ⊗−1
j=−` ϑm+j+1 (τE
(Bj )) ⊗ ϑm+1 (B0 )
r
EY
×e−isLm+1 e−iτ Lm · · · e−iτ Lp+1 ⊗pi=1 ei[(m−i)τ +s]LE Ai Ω
hBj iΩE ,
j=1
(4.12)
BJM November 4, 2005
22
where we have used that Ω is in the kernel of the Kj , and where we have commuted
the product of the freely evolved Aj ’s to the right through the propagators which act
on different factors.
Because the system E has the property of return to equilibrium the propagator
i[(m(t)−i)τ
+s(t)]LE converges to the projection P
e
ΩE = |ΩE ihΩE |, as t → ∞, in the weak
sense on HE . Define the orthogonal projection Q⊥
p = 1l − Qp on H, where Qp =
⊗pj=1 ϑj (PΩE ). We want to show that
eN eiτ K1 · · · eiτ Kp eiτ Lp+1 · · · eiτ Lm eisLm+1
lim Ω, (B 0 )∗ B 0 U
t→∞
(−j−1)τ +s
×AS ⊗−1
j=−` ϑm+j+1 (τE
(Bj )) ⊗ ϑm+1 (B0 )
E
p
i[(m−i)τ +s]LE
×e−isLm+1 e−iτ Lm · · · e−iτ Lp+1 Q⊥
Ai Ω = 0.
p ⊗i=1 e
(4.13)
To do so we split the Hilbert space as H = H1 ⊗ H2 , where H1 = HS ⊗m≥p+1 HE , and
H2 = ⊗pi=1 HE , and set
(−j−1)τ +s
ψ1m = eiτ Lp+1 · · · eiτ Lm eisLm+1 AS ⊗−1
(Bj ))
j=−` ϑm+j+1 (τE
−isLm+1 −iτ Lm
−iτ Lp+1
⊗ϑm+1 (B0 ) e
e
···e
ΩS ⊗m≥p+1 ΩE ∈ H1 ,
p
i[(m−i)τ +s]LE
Ai ⊗pi=1 ΩE ∈ H2 .
ψ2m = Q⊥
p ⊗i=1 e
⊥
Q⊥
p is a sum of terms, each one containing the operator PΩE = 1lE − PΩE acting on at
m(t)
least one of the p factors in H2 . Consequently we have ψ2
→ 0, weakly in H2 , as
m(t)
m(t)
m(t)
t → ∞. Since ψ1
is uniformly bounded in t it follows that ψ1
⊗ ψ2
converges
weakly to zero in H, as t → ∞. This proves relation (4.13).
Thus, in the limit t → ∞, the only contribution to (4.12) comes from the part
where all the free propagators ei[(m−i)τ +s]LE are replaced by PΩE . This shows that
eN eiτ K1 · · · eiτ Kp eiτ Lp+1 · · · eiτ Lm eisLm+1
lim E(t) − Ω, (B 0 )∗ B 0 U
t→∞
(−j−1)τ +s
×AS ⊗−1
j=−` ϑm+j+1 (τE
×e
−isLm+1 −iτ Lm
e
···e
−iτ Lp+1
Ω
(Bj )) ⊗ ϑm+1 (B0 )
p
Y
i=1
hAi iΩE
r
Y
hBj iΩE = 0.
(4.14)
j=1
We may now, as we did above, turn the operators Lj in (4.14) into Kj ’s, also for the
remaining indices j = p + 1, . . . , m + 1, to arrive at
eN eiτ K1 · · · eiτ KN PN eiτ KN +1 · · · eiτ Km eisKm+1
lim E(t) − Ω, (B 0 )∗ B 0 U
t→∞
(−j−1)τ +s
×AS ⊗−1
(Bj )) ⊗ ϑm+1 (B0 )Ω
j=−` ϑm+j+1 (τE
p
r
Y
Y
×
hAi iΩE
hBj iΩE = 0,
(4.15)
i=1
j=1
where we introduce the projection PN = ⊗m≥N +1 ϑm (PΩE ) (that projection comes from
eN and through the first
the left factor of the inner product and slips through (B 0 )∗ B 0 U
N propagators).
BJM November 4, 2005
23
We have
ϑm+1 (PΩE )eisKm+1 AS ⊗ ϑm+1 (B0 ) Ω = D0 (s)Ω,
(4.16)
where D0 (s) is a linear operator acting nontrivially only on HS . D0 (s) depends on AS ,
B0 , the interaction V and s = s(t), but it is independent of m = m(t). In the same
way we define D1 (s) ∈ B(HS ) by
ϑm (PΩE )eiτ Km ϑm (τEs (B−1 ))D0 (s)Ω = D1 (s)D0 (s)Ω,
(4.17)
and then D2 (s), . . . , D` (s) ∈ B(HS ). Hence the inner product in (4.15) can be written
as
D
E
eN eiτ K1 · · · eiτ KN PN eiτ KN +1 · · · eiτ Km−` D(s)Ω ,
Ω, (B 0 )∗ B 0 U
(4.18)
where D(s) = D` (s) · · · D0 (s) ∈ B(HS ). Since Ω = P Ω, where P is the projection onto
ΩC , c.f. (1.8), we have
PN eiτ KN +1 · · · eiτ Km−` D(s)Ω = P eiτ KN +1 · · · eiτ Km−` P D(s)Ω.
(4.19)
The reduced product of the propagators on the r.h.s. is the product of the reduced
propagators, as we show in the following proposition.
Proposition 4.1 For any q ≥ 1, let t1 , . . . tq ∈ R, and let m1 , . . . , mq ≥ 1 be distinct
integers. Then we have
P eit1 Km1 · · · eitq Kmq P = P eit1 Km1 P · · · P eitq Kmq P.
(4.20)
Proof of Proposition 4.1.
If Q is a projection we set Q⊥ = 1l − Q. We have
P ⊥ eitq Kmq P ∈ Ran ϑmq (PE⊥ ) since on all factors of HC with label m 6= mq the projection
ϑm (PE ) coming from P commutes with eitq Kmq . It follows that P eit1 Km1 · · · P ⊥ eitq Kmq P =
P eit1 Km1 · · · ϑmq (PE⊥ )P ⊥ eitq Kmq P = 0, because ϑmp (PE⊥ ) can be commuted to the left
to hit P . Thus we have P eit1 Km1 · · · eitq Kmq P = P eit1 Km1 · · · eitq−1 Kmq−1 P eitq Kmq P and
we can repeat the argument. This proves the proposition.
According to (4.20) the r.h.s. of (4.19) equals M m−`−N D(s)Ω, where M = P M P
is the operator introduced in (2.18). We then obtain the following result from (4.19),
(4.18) and (4.15):
D
E
eN eiτ K1 · · · eiτ KN P M m(t)−`−N P D(s(t))Ω
lim E(t) − Ω, (B 0 )∗ B 0 U
t→∞
×
p
Y
i=1
hAi iΩE
r
Y
hBj iΩE = 0.
(4.21)
j=1
In order to further simplify the scalar product in (4.21) we use the ergodicity assumption
(E). We have M m(t)−`−N → π = |ΩS ihΩ∗S |, as t → ∞, in the topology of B(HS ) (c.f.
(2.19)), and since D(s(t)) is uniformly bounded in t we obtain
p
r
Y
Y
∗
0 ∗ 0
lim E(t) − Ω, (B ) B Ω hΩS , P D(s(t))P ΩS i
hAi iΩE
hBj iΩE = 0.
t→∞
i=1
j=1
(4.22)
BJM November 4, 2005
24
Remember that ψ is approximated by B 0 Ω, i.e., given an arbitrary ε > 0 we choose B 0
s.t. kψ − B 0 Ωk ≤ ε. Thus
Ω, (B 0 )∗ B 0 Ω = kB 0 Ωk2 = (kψk + O(ε))2 = 1 + O(ε),
(4.23)
which we can use in (4.22) to arrive at
p
r
Y
Y
∗
lim E(t) − hΩS , P D(s(t))P ΩS i
hAi iΩE
hBj iΩE = 0.
t→∞
i=1
(4.24)
j=1
Note that if 1 is a degenerate eigenvalue of M , (4.22) becomes
p
r
Y
Y
lim E(t) − Ω, (B 0 )∗ B 0 πP D(s(t))P ΩS
hAi iΩE
hBj iΩE
t→∞
i=1
= 0.
(4.25)
j=1
Thus, E(t) has a τ −periodic asymptotic behaviour (s(t) is periodic) but which depends
a priori on B 0 and thus on the initial state ω: the simplification due to (4.23) does not
hold anymore.
Finally by the definition of D(s), c.f. (4.18), (4.17), (4.14) we get
hΩ∗S , P D(s(t))P ΩS i
(4.26)
D
E
(`−1)τ +s
∗
iτ K1
iτ K` isK`+1
s
=
ΩS , P e
···e
e
AS ⊗ τ E
(B−` ) ⊗ · · · ⊗ τE (B−1 ) ⊗ B0 P ΩS
Using the invariance P ΩS = e−isK`+1 e−iτ K` · · · e−iτ K1 P ΩS and passing from K’s to L’s
we obtain
hΩ∗S , P D(s(t))P ΩS i
D
(`−1)τ +s
=
Ω∗S , P eiτ L1 · · · eiτ L` eisL`+1 AS ⊗ τE
(B−` ) ⊗ · · ·
E
· · · ⊗ τEs (B−1 ) ⊗ B0 e−isL`+1 e−iτ L` · · · e−iτ L1 P ΩS
D
=
Ω∗S , P eiτ L1 · · · eiτ L` eisL`+1 (U`− )∗ AS ⊗ B−` ⊗ · · · ⊗ B0 U`−
E
×e−isL`+1 e−iτ L` · · · e−iτ L1 P ΩS
D
E
`τ +s(t)
∗
=
ΩS , P αRI
(AS ⊗ B−` ⊗ · · · ⊗ B0 ) P ΩS .
(4.27)
We conclude the proof of (2.29), (2.30) and hence the proof of Theorem 2.4 by plugging
(4.27) into (4.24).
4.2
Proof of Theorem 2.5
The proof is a simple modification of the proof of Theorem 2.4. We indicate the main
steps. Proceeding as in Section 4.1 we see that (compare with (4.12))
D
E
eN eiτ K1 · · · eiτ Km eisKm+1 AS ⊗ ϑm+1 (B0 )Ω , (4.28)
C(t; A, AS , B0 ) = Ω, (B 0 )∗ B 0 A U
BJM November 4, 2005
25
where we may assume that A acts trivially on factors of HC with index > N . As in Section 4.1 we replace the product of the propagators in (4.28) by eiτ K1 · · · eiτ KN M m−N −1 .
Taking the limit m = m(t) → ∞ then yields
s(t)
lim C(t; A, AS , B0 ) − Ω, (B 0 )∗ B 0 AΩ ω+ P αRI (AS ⊗ B0 )P = 0,
(4.29)
t→∞
with a speed of convergence dictated by Proposition 2.2. To complete the proof of
Theorem 2.5 we notice that hΩ, (B 0 )∗ B 0 AΩi = hAiB 0 Ω = ω(A) + O(ε), for arbitrary
ε > 0 (c.f. the argument before (4.23)).
4.3
Proof of Proposition 2.1
Let Km , Lm be the operators on H that act trivially on all factors except on HS times
the m–th HE , where they act as K, L. Given any t ∈ R, A, B ∈ MS (⊂ M) we have
BΩS ⊗ ΩC , eitL1 · · · eitLm Ae−itLm · · · e−itL1 ΩS ⊗ ΩC H
= BΩS ⊗ ΩC , eitK1 · · · eitKm AΩS ⊗ ΩC H
= BΩS , (P eitK P )m AΩS H ,
(4.30)
S
where we use that K and L implement the same dynamics on M and (2.16) in the first
step, and Proposition 4.1 in the second step. Since BΩS is dense in HS it follows from
(4.30) that
k(P eitK P )m AΩS k ≤ kAk.
(4.31)
So far we have not used that dim HS < ∞. In the finite-dimensional case, MS ΩS is
not only dense in HS , but for any ψ ∈ HS there exists an A ∈ MS s.t. ψ = AΩS . Thus
(4.31) and the uniform boundedness principle give that
k(P eitK P )m k < ∞.
sup
(4.32)
t∈R,m≥0
The facts that the spectrum of P eitK P lies in the unit disk in C, and that all eigenvalues on the unit circle must be semisimple follow from the uniform boundedness of
k(P eitK P )m k in m. They can be shown using an easy Jordan canonical form argument.
4.4
Proof of Proposition 2.6
Given any normal state ω of M and any unitary U on H we have the following relation
[7] (our definition of entropy differs from the one in [7] by a sign)
Ent ω(U ∗ · U )|ω0 − Ent(ω|ω0 )
h X
i
X
= ω U ∗ βE
LE,k + βS LS U − βE
LE,k − βS LS .
(4.33)
k
k
In case U is a dynamics of the system, (4.33) is called the entropy production formula.
We take U = URI (t). The argument of ω in (4.33) can be written as
n
X
X
o
t
t
βE αRI
LE,k + LS −
LE,k − LS − (βE − βS ) αRI
(LS ) − LS ,
k
k
BJM November 4, 2005
26
so it suffices to prove (2.43) for βS = βE = β. We want to show that
X
X
t
t
αRI
LE,k + LS −
LE,k − LS = ∆E(t) − αRI
(Vm(t)+1 ) + V1 .
k
(4.34)
k
It is clear that the sums in the l.h.s. of (4.34) extend only from k = 1 to k = m + 1.
We examine the difference of the first two terms on the l.h.s., for k = m + 1. With
t = mτ + s, we obtain for this difference the expression
t
t
αRI
(LE,m+1 + LS + Vm+1 ) − αRI
(Vm+1 ) − LE,m+1
mτ
t
= αRI
(LE,m+1 + LS + Vm+1 ) − αRI
(Vm+1 ) − LE,m+1
mτ
mτ
t
= αRI
(LS ) + αRI
(Vm+1 ) − αRI
(Vm+1 )
mτ
mτ
t
= αRI
(LS + Vm ) + αRI
(Vm+1 − Vm ) − αRI
(Vm+1 )
mτ
t
= αRI
(LS + Vm ) + j(m) − αRI
(Vm+1 ),
(4.35)
mτ (L
where we use in the second step αRI
E,m+1 ) = LE,m+1 , and in the last step we use
t (L
mτ
definition (2.35). We now add to (4.35) the expression αRI
E,m )−LE,m = αRI (LE,m )−
LE,m (i.e., the term with k = m in the sums of the l.h.s. of (4.34)) and repeat the
manipulations leading to (4.35) to obtain for this sum the expression
(m−1)τ
αRI
t
(LS + Vm−1 ) + j(m − 1) + j(m) − αRI
(Vm+1 ).
It is now clear how to continue this process until all terms with k = 1, . . . , m + 1 in the
sums of the l.h.s. of (4.34) are taken care of. We obtain
m(t)
l.h.s. of (4.34) =
X
t
j(k) + V1 − αRI
(Vm(t)+1 ).
(4.36)
k=1
Definition (2.41) yields the result
(2.43).
t |ω
To see that Ent ω ◦ αRI
0 is continuous in t ≥ 0 we show that
(m−1)τ +s0
mτ +s
|ω0 − Ent ω ◦ αRI
Ent ω ◦ αRI
|ω0
(4.37)
converges to zero, for all m ≥ 1, as s ↓ 0 and s0 ↑ τ . Expression (2.43) yields
(4.37) = ω βE ∆E(mτ + s) − ∆E((m − 1)τ + s0 )
(4.38)
(m−1)τ +s0
mτ +s
−αRI
βE Vm+1 + ∆βLS + αRI
βE Vm + ∆βLS ,
where ∆β = βE − βS . The first argument of ω in (4.38) equals βE j(m), see (2.41).
Next,
(m−1)τ +s0
mτ +s
mτ
αRI
(Vm+1 ) − αRI
(Vm ) −→ αRI
(Vm+1 − Vm ),
(4.39)
as s ↓ 0, s0 ↑ τ . But the r.h.s. of (4.39) is just j(m), see (2.35). This will compensate
mτ +s
the contribution of the first argument in ω of (4.38) in the limit. Finally, αRI
(LS ) −
(m−1)τ +s0
0
αRI
(LS ) tends to zero, as s ↓ 0, s ↑ τ . Consequently, (4.37) vanishes in the
limit.
BJM November 4, 2005
4.5
27
Proof of Proposition 2.7
By (2.43) and (2.41), the change of entropy within an interval of duration τ is given by
t+τ
t
Ent(ω ◦ αRI
|ω0 ) − Ent(ω ◦ αRI
|ω0 )
=
(4.40)
t+τ
t
βE ω j(m + 1) − αRI
(Vm+2 ) + αRI
(Vm+1 )
t+τ
t
−(βE − βS ) ω αRI (LS ) − αRI (LS ) ,
(4.41)
(4.42)
where j(k) is given in (2.35), and m = m(t). Taking into account (2.35), (2.38) and
Theorem 2.3 we see that the limit t → ∞ of (4.41) is βE ω+ (j+ ). We claim that (4.42)
vanishes as t → ∞. This can be seen in the following way. An application of Theorem
2.4 shows that
Z τ
t+τ
t
s
lim ω αRI (LS ) − αRI (LS ) =
ω+ (P αRI
(i[V, LS ])P )ds,
(4.43)
t→∞
0
where we use that [V, LS ] ∈ MS ⊗ ME , which follows from the fact that eitLS V e−itLS ∈
MS ⊗ ME for all t ∈ R. Therefore
Z τ
t+τ
t
s
(i[V, LS ])P )ds
Ent(ω ◦αRI |ω0 )−Ent(ω ◦αRI |ω0 ) −→ βE ω+ (j+ )−(βE −βS ) ω+ (P αRI
0
(4.44)
as t → ∞. On the other hand, (2.46) shows that
βE
1
t+τ
t
|ω0 ) − Ent(ω ◦ αRI
|ω0 ) −→
Ent(ω ◦ αRI
ω+ (j+ ),
t
τ
(4.45)
in the limit t → ∞. Then the following general fact proves that (4.43) must be equal
to zero:
If a locally bounded function f on R has the property f (t+τ )−f (t) → a, as t → ∞,
for some a ∈ R and some τ > 0, then f (t)/t → a/τ , as t → ∞.
This concludes the proof of Proposition (2.7).
4.6
Proof of Theorem 3.1
Using a Dyson expansion we get
Z τ
iτ Kλ
iτ K0
e
= e
+ iλ
dtei(τ −t)K0 W eitK0
(4.46)
0
Z τZ t
−λ2
ei(τ −t)K0 W ei(t−s)K0 W eisK0 ds dt + O(λ3 ).
0
0
BJM November 4, 2005
28
Inserting (3.5)-(3.14)-(3.16)-(3.17) in (4.46), one gets after a somewhat lengthy but
straightforward computation
2
Mλ = eiτ LS + iλ sin(τ ) e−βh/2 gβ σx ⊗ e−iτ σz − eiτ σz ⊗ σx
h
Z τ Z t Z Z
kgβ (r1 )k2 kgβ (r2 )k2 e−βr1 ei(t−s)(r2 −r1 ) dr1 dr2
+λ2
0 0
R+ R+
+ke−βh/2 gβ k4h ei(τ −t)σz σx eitσz ⊗ e−i(τ −s)σz σx e−isσz
−ei(τ −2t+2s)σz ⊗ e−iτ σz ds dt
Z τ Z t Z Z
2
kgβ (r1 )k2 kgβ (r2 )k2 e−βr2 ei(t−s)(r2 −r1 ) dr1 dr2
+λ
+
+
0 0
R R
+ke−βh/2 gβ k4h ei(τ −s)σz σx eisσz ⊗ e−i(τ −t)σz σx e−itσz
−eiτ σz ⊗ e−i(τ −2t+2s)σz ds dt
+O(λ3 ).
(4.47)
Using perturbation theory [8], we then find that Mλ has 4 distinct eigenvalues: 1, e0 (λ),
e+ (λ), e− (λ) which are given by
e0 (λ) = 1 − λ2 τ 2 (α1 + α2 ) + O(λ3 ),
(4.48)
λ2 τ 2
1−
(α1 + α2 ) + iλ2 τ ke−βh/2 gβ k4h
(4.49)
2
ZZ
−
(e−βr1 + e−βr2 )kgβ (r1 )k2g kgβ (r2 )k2g
1 − sinc(τ (r1 − r2 − 2)) ×
+ O(λ3 ),
r1 − r2 − 2
e+ (λ) = e
2iτ
e− (λ) = e
−2iτ
λ2 τ 2
(α1 + α2 ) − iλ2 τ ke−βh/2 gβ k4h
(4.50)
1−
2
ZZ
−
(e−βr1 + e−βr2 )kgβ (r1 )k2g kgβ (r2 )k2g
1 − sinc(τ (r1 − r2 − 2)) ×
+ O(λ3 ),
r1 − r2 − 2
where the αj ’s are defined in (3.21). Since they are strictly positive numbers, this
proves that for |λ| small enough, the operator Mλ satisfies Assumption (E).
It remains to prove that the asymptotic state ω+,λ is indeed given by (3.20). For
that purpose, it suffices to compute the eigenvector Ω∗S (λ) of Mλ∗ for the eigenvalue 1,
which is obtained by perturbation theory:
√
√
α1 2
α2 2
∗
ΩS (λ) =
ψ11 +
ψ22
(4.51)
α1 + α2
α1 + α2
α1 − α2
+λke−βh/2 gβ k2 √
(ψ12 + ψ21 ) + O(λ2 ).
2(α1 + α2 )
BJM November 4, 2005
4.7
29
Proof of Theorem 3.2
Let P0 denote the spectral projection on the kernel of L0 , and Ω∗0 the main term in
the expansion of Ω∗S (λ), i.e. Ω∗S (λ) = Ω∗0 + O(λ) (see e.g. (4.51)). Then, using (2.23),
(2.38) and perturbation theory, one gets
Z τ
2
hΩ∗0 ⊗ ΩE |V (e−itL0 − eitL0 )(1 − P )V ΩS ⊗ ΩE i
ω+ (j+ ) = λ i
0
Z τ
∗
itL0
hΩ0 ⊗ ΩE |W P0 e V ΩS ⊗ ΩE i + O(λ3 ). (4.52)
−i
0
Moreover, as a general fact, one has P0 Ω∗0 ⊗ ΩE = Ω∗0 ⊗ ΩE . Now, the quadratic
interaction satisfies P0 W P0 = 0. Hence the second term on the right hand side of
(4.52) cancels.
Using (2.50),(3.16) and (4.51) we thus have
Z
0
λ2 βτ
2
0 2
0
2 τ (2 − r + r )
dS+ =
kgβ (r)k kgβ (r )k (2 − r + r )sinc
α1 + α2
2
0
×(α2 e−βr − α1 e−βr )drdr0 + O(λ3 ).
Inserting the expression for αj in the integral, one finally has
Z
λ2 βτ
dS+ =
kgβ (r1 )k2 kgβ (r2 )k2 kgβ (r3 )k2 kgβ (r4 )k2
(4.53)
2(α1 + α2 )
2 τ (2 − r3 + r4 )
2 τ (2 − r1 + r2 )
sinc
×sinc
2
2
×(r2 + r3 − r1 − r4 ) e−β(r1 +r4 ) − e−β(r2 +r3 ) + O(λ3 ),
where the integral runs over the four variables rj ∈ R+ . The result follows then from
the fact that, for any real x and y, (x − y)(e−βy − e−βx ) is non-negative.
4.8
Proof of Theorem 3.3
The proof goes exactly in the same way as for Theorems 3.1 and 3.2. We just give
the expressions for the eigenvalues (different from 1) of Mλ as well as the one for the
entropy production.
e0 (λ) = 1 − λ2 τ 2 (α1 + α2 ) + O(λ4 ),
(4.54)
Z
λ2 τ 2
e+ (λ) = e2iτ 1 −
(α1 + α2 ) + iλ2 τ 2 kg(r)k2g
(4.55)
2
1 − sinc(τ (2 − r)) 1 − sinc(τ (2 + r))
×
+
+ O(λ4 ),
2−r
2+r
e+ (λ) = e
−2iτ
Z
λ2 τ 2
2 2
1−
(α1 + α2 ) − iλ τ
kg(r)k2g
(4.56)
2
1 − sinc(τ (2 − r)) 1 − sinc(τ (2 + r))
×
+
+ O(λ4 ),
2−r
2+r
BJM November 4, 2005
30
λ2 βτ
α1 + α2
τ (r − 2)
kgβ (r)k2g kgβ (r0 )k2g sinc2
2
0
τ (r + 2)
0
sinc2
(r + r0 )(1 − e−β(r+r ) )
2
Z
2
λ βτ
0
+
kgβ (r)k2g kgβ (r0 )k2g (r0 − r)(e−βr − e−βr )
2(α1 + α2 )
0
2 τ (r − 2)
2 τ (r − 2)
× sinc
sinc
2
2
0
2 τ (r + 2)
2 τ (r + 2)
sinc
+ O(λ3 ).
+sinc
2
2
dS+ =
Z
(4.57)
All the integrals run over R+ , and in the formula for the entropy the integration is
computed w.r.t. both r and r0 .
4.9
Proof of Theorem 3.4
Once again the proof goes the same way, and we only give the expressions for the
eigenvalues (different from 1) of Mλ and for the entropy production.
e0 (λ) = 1 −
λ2 τ 2
(α1 + α2 ) + O(λ4 ),
1 + e−βEE
e+ (λ) = e
iτ ES
1−
λ2 τ 2
α1 + α2 + (1 + e−βEE )sinc2
2(1 + e−βEE )
τ EE
2
×(|a|2 + |d|2 − ād − ad̄)
+i
+O(λ4 ),
λ2 τ 2
1 + e−βEE
1 − sinc(τ EE )
(|a|2 − |d|2 )
τ EE
ād − ad̄
−βEE
2 τ EE
+(1 − e
)sinc
2
2i
1 − sinc(τ (EE − ES )) 2
−(1 + e−βEE )
|b|
τ (EE − ES )
−βEE 1 − sinc(τ (EE + ES ))
2
+(1 + e
)
|c|
τ (EE + ES )
(1 − e−βEE )
BJM November 4, 2005
e− (λ) = e
−iτ ES
31
1−
λ2 τ 2
α1 + α2 + (1 + e−βEE )sinc2
2(1 + e−βEE )
τ EE
2
×(|a|2 + |d|2 − ād − ad̄)
λ2 τ 2
+i
1 + e−βEE
1 − sinc(τ EE )
(|d|2 − |a|2 )
τ EE
τ EE ad̄ − ād
+(1 − e−βEE )sinc2
2
2i
1 − sinc(τ (EE − ES )) 2
+(1 + e−βEE )
|b|
τ (EE − ES )
1 − sinc(τ (EE + ES )) 2
−(1 + e−βEE )
|c|
τ (EE + ES )
(1 − e−βEE )
+O(λ4 ),
dS+ =
λ2 βτ EE (1 − e−βEE )
×
(α1 + α2 )(1 + e−βEE )
2
2
−βEE
2
2 τ (EE − ES )
2 τ EE
|b| (|a| + e
|d| )sinc
sinc
2
2
2 −βEE
2
2
2 τ (EE + ES )
2 τ EE
+|c| (e
|a| + |d| )sinc
sinc
2
2
τ (EE + ES )
+2|b|2 |c|2 (1 + e−βEE )sinc2
2
2 τ (EE + ES )
× sinc
+ O(λ3 ).
2
Acknowledgements. This work was initiated during a visit of M.M. to the
Institut Fourier of Université de Grenoble I while L.B. was a postdoc there, and it was
essentially completed during a visit of L.B. to the Mathematics Department of McGill
University. We are grateful to the Institut Fourier for financial support and to both
the Institut Fourier and McGill University for their hospitality.
References
[1] Araki, H., Wyss, W.: Representations of canonical anticommutation relations
Helv. Phys. Acta 37 136–159 (1964).
[2] Attal, S., Joye, A.: Weak Coupling and Continuous Limits for Repeated Quantum
Interactions preprint mp-arc 05-5.
BJM November 4, 2005
32
[3] Attal, S., Pautrat, Y.: From repeated to continuous quantum interactions. To
appear.
[4] Wellens, T., Buchleitner, A., Kümmerer, B., Maassen, H.: Quantum State Preparation via Asymptotic Completeness. Phys. Rev. Lett. 85, no.16, 3361–3364 (2000)
[5] Bratteli, O., Robinson, D.W., Operator Algebras and Quantum Statistical Mechanics I, II. Texts and Monographs in Physics, Springer-Verlag, 1987.
[6] Jaks̆ić, V., Pillet, C.-A.: Non-equilibrium steady states of finite quantum systems
coupled to thermal reservoirs. Comm. Math. Phys. 226, no.1, 131–162 (2002).
[7] Jaks̆ić, V., Pillet, C.-A.: A note on the entropy production formula. Advances in
differential equations and mathematical physics (Birmingham, AL, 2002), 175–180,
Contemp. Math., 327, Amer. Math. Soc., Providence, RI, 2003.
[8] Kato, K., Perturbation Theory for Linear Operators. 2nd edition. Springer, Berlin,
1976.
[9] Merkli, M., Mück, M., Sigal, I.M.: Instability of Equilibrium States for Coupled
Heat Reservoirs at Different Temperatures. submitted.
[10] Pillet, C.-A., Quantum Dynamical Systems. Springer Lecture Notes in Mathematics, to appear
[11] Vogel, K., Akulin, V.M., Schleich, W.P.: Quantum State Engineering of the Radiation Field. Phys. Rev. Lett. 71, no.12, 1816–1819 (1993)